Scale-Sensitive IOU Loss: An Improved Regression Loss Function in Remote Sensing Object Detection

Shuangjiang Du,Baofu Zhang,Pin Zhang

doi:10.1109/access.2021.3119562

Shuangjiang Du, Baofu Zhang + Show 1 more

Open Access

https://doi.org/10.1109/access.2021.3119562

Copy DOI

Abstract

Regression loss function in object detection model plays an important factor during training procedure. The IoU based loss functions, such as CIOU loss, achieve remarkable performance, but still have some inherent shortages that may cause slow convergence speed. The paper proposes a Scale-Sensitive IOU(SIOU) loss for the object detection in multi-scale targets, especially the remote sensing images to solve the problem where the gradients of current loss functions tend to be smooth and cannot distinguish some special bounding boxes during training procedure in multi-scale object detection, which may cause unreasonable loss value calculation and impact the convergence speed. A new geometric factor affecting the loss value calculation, namely area difference, is introduced to extend the existing three factors in CIOU loss; By introducing an area regulatory factor $\gamma $ to the loss function, it could adjust the loss values of the bounding boxes and distinguish different boxes quantitatively. Furthermore, we also apply our SIOU loss to the oriented bounding box detection and get better optimization. Through extensive experiments, the detection accuracies of YOLOv4, Faster R-CNN and SSD with SIOU loss improve much more than the previous loss functions on two horizontal bounding box datasets, i.e, NWPU VHR-10 and DIOR, and on the oriented bounding box dataset, DOTA, which are all remote sensing datasets. Therefore, the proposed loss function has the state-of-the-art performance on multi-scale object detection.

Highlights

IntroductionR EGRESSION loss function is a significant factor that affects the object detection performance, the n norm is first used to calculate regression loss of which Smooth L1norm [14] is an improvement
R EGRESSION loss function is a significant factor that affects the object detection performance, the n norm is first used to calculate regression loss of which Smooth L1norm [14] is an improvement.The IoU based loss loss functions are the widely used regression loss functions in many object detection models, of which the first proposed is IOU loss [15] and it performs better than the former in many datasets
IOU loss has some inherent disadvantages especially when the bounding box do not overlap with the ground truth box, that is, IoU values is 0, and the Generalized IOU (GIOU) loss [16] improves the IOU loss

Summary

Introduction

R EGRESSION loss function is a significant factor that affects the object detection performance, the n norm is first used to calculate regression loss of which Smooth L1norm [14] is an improvement. The IoU based loss loss functions are the widely used regression loss functions in many object detection models, of which the first proposed is IOU loss [15] and it performs better than the former in many datasets. Distance IOU(DIOU) loss and Complete IOU(CIOU) loss [17] are proposed, arguing that the former two loss functions still have some lacks in theory. In CIOU loss, it summarizes three geometric factors that affect the regression loss value calculation, namely overlap area, center point distance and aspect ratio. Efficient IOU loss [18] combines the theory of Focus Loss [19] and add hard example mining mechanism into CIOU loss, which improves the performance of the later one

Objectives

Methods

Results

Conclusion